Vous êtes sur la page 1sur 8

48 s Transportation Research Record 1857

Paper No. 03- 3697


A new transit operating strategy is presented in which service vehicles
operate in pairs with the lead vehicle providing an all-stop local service
and the following vehicle being allowed to skip some stops as an express
service. The underlying scheduling problem is formulated as a nonlinear
integer programming problem with the objective of minimizing the total
costs for both operators and passengers. A sensitivity analysis using a
real-life example is performed to identify the conditions under which the
proposed operating strategy is most advantageous.
The design and operations of public transit services involve many
complicated issues. On the one hand, transit routes are expected to
have a xed schedule or service interval (headway) and to provide
broad coverage. On the other hand, the origin-destination (O-D) pas-
senger demands along a route often vary signicantly, with some
stops, such as those downtown, having a relatively large number of
passengers boarding and alighting and others having few passengers.
Such an O-D pattern cannot be efficiently serviced by transit with
xed routes and schedules. Balancing the needs of delivering reliable
and consistent services and the costs of providing such services has
continuously been one of the major challenges facing the transit
agencies.
In the past 30 years, many operating strategies have been proposed
with the goal of better matching services covered by transit routes
with the O-D demand distribution along the routes. One of the central
ideas has been exible routing and scheduling that integrates express
services (stop only at a few stops) with local services (stop at all
stops). Flexible routing and scheduling strategies can be classied into
four general categories (1): (a) zone scheduling, including restricted
zonal service, semirestricted zonal service, and limited-stop zonal ser-
vice; (b) short turning; (c) deadheading; and (d) dynamic stop skip-
ping. In zone scheduling, the whole route is divided into several
zones. The inbound buses make all stops within a single zone and then
run without stopping to the terminus, while outbound vehicles oper-
ate in the reverse manner. With this operating strategy, passenger
travel time and the required numbers of vehicles and drivers may be
reduced. This operating strategy, however, has two disadvantages,
including reduced service frequency and increased waiting time and
the possible requirement of cross-zone transfer. Turnquist (2, 3) was
among the rst to study the zone scheduling method with the objec-
tive of simultaneously determining the optimal zone division and
vehicle allocation. This model was extended by Jordan and Turnquist
(4) to consider both the mean and variance of passengers trip time.
Short turning (58) strategies consist of a system of short-turn and
full-length trips operating along the same route. This express service
is particularly suitable for routes in which the O-D demand peaks
in a specific zone and decreases substantially outside that zone.
The short-turn trips cover only the high-demand zone while the
full-length trips run the whole route. Compared with a local ser-
vice consisting only of full-length trips, short-turn operations
require fewer full-length trips and thus fewer service vehicles. The
critical design issue in short-turn services is to determine the turn-
back point and the route schedule to balance passenger loads
among the trips and to minimize the total fleet size and passenger
wait time.
Deadheading, another type of express service, involves schedul-
ing some service vehicles to run empty through a number of stations
at the beginning or the end of their routes to save time and hence
reduce the headways at later stations. Furth and Day (1) and Furth
(9) studied the preplanned deadheading problem, which was for-
mulated to minimize the eet size required to meet a regular alter-
nating deadheading schedule. More recently, Eberlein (10) and
Eberlein et al. (11) investigated the deadheading problem in the con-
text of real-time transit controls. The objective was to determine
which vehicle to deadhead and at which stations. Both problems
were formulated as nonlinear quadratic programs.
Dynamic stop skipping, also called expressing, is an operating
strategy that has been frequently used in heavy-demand corridors.
The basic idea behind this control strategy is to allow those vehicles
that are late and behind schedule to skip certain low-demand stops
and increase operating speed. In contrast to other strategies, which
are mostly off-line strategies and designed at the service planning
stage, dynamic stop skipping is an on-line strategy determined in real
time. One disadvantage of this control strategy is that passengers
with either their origin or destination stop being skipped have to wait
for at least another headway to get service.
Eberlein (10) formulated the stop-skipping problem as an integer
nonlinear programming model (INLP) with both quadratic objective
function and constraints. Li and Wu (12), Li et al. (13), and Li (14)
focused on the application scenario of a heavily used transit system
with short headway (e.g. 2 to 3 min) and proposed two alternative
models: a deterministic mixed INLP that assumed known average
travel time and demands but variable dwell times, and a stochastic
integer program with random travel times and demands but deter-
ministic dwell times. Araya et al. (15) proposed a method for the gen-
eration of optimal schedules for on-line train traffic control. Lin et al.
(16) investigated the combined strategy of stop skipping and holding
and suggested that tight controls increase passenger travel times and
therefore should be avoided. In their approach, a minimum headway
could not be guaranteed, especially at some O-D pairs with low travel
demand, because no limitation was imposed on dispatching patterns.
The solutions of Eberlein (10), Li and Wu (12), and Li (14) were
based on either very simplied formulation or heuristic algorithms. Li
et al. (13) and Li (14) compared different models and algorithms and
Real-Time Optimization Model for Dynamic
Scheduling of Transit Operations
Liping Fu, Qing Liu, and Paul Calamai
L. Fu, Department of Civil Engineering, and Q. Liu and P. Calamai, Department
of Systems Design Engineering, University of Waterloo, Waterloo, Ontario N2L
3G1, Canada.
concluded that solutions from controlling a small subset of vehicles
were very close to those when all vehicles were considered.
In what follows, a new dynamic scheduling strategy is presented
that aims to strike an optimal balance between the benets to opera-
tors and those to passengers. In the second section, a mathematical
model is developed for the dynamic scheduling problem. Results of a
sensitivity analysis are presented in the third section. The last section
highlights the conclusions and suggests future research directions.
MODEL FORMULATION
System Denition
Consider a transit route consisting of N stops labeled sequentially
from the terminal (Stop 1) to the end of the route (Stop N), as indi-
cated in Figure 1. Buses are dispatched at the terminal according to
a given schedule or headway. The dispatching functions are assumed
by a dispatch center, which is equipped with a computer-aided dis-
patching system and an automatic vehicle location system. At any
point of time, a service vehicle could be in one of the following three
possible states: moving in-between stops, dwelling at a stop for pas-
senger alighting and boarding, or waiting to be dispatched at the ter-
minal. The problem this paper focuses on arises whenever a vehicle
is in the last statethat is, it is waiting to be dispatched at the termi-
nal and requiring instructions on which stops to make on its route.
This problem is commonly referred to as the dynamic transit sched-
uling problem or the dynamic stop-skipping problem.
The common approach to the dynamic scheduling problem has
been to apply stop skipping controls to every vehicle dispatched
from the terminal. This approach, seemingly optimal in a theoretical
sense, suffers a critical limitation. That is, transit users cannot be
guaranteed a minimum level of service as the model could produce
solutions that allow several consecutive buses to skip the same stops.
In this study, a new stop-skipping strategy is proposed, in which only
every other bus will be allowed to express and skip stops. In other
words, the stop-skipping control will be applied to alternate buses.
One immediate consequence of this strategy is that the schedule head-
way at any stop will be no more than twice the dispatch headway. For
example, if the scheduled dispatch headway of a route is 5 min, the
maximum scheduled headway at any stop on the route will be 10 min.
In addition, the skipping pattern set at the terminal is xed once the
Fu et al. Paper No. 03- 3697 49
bus departs from the terminal. This control strategy has two advan-
tages. First, a minimum level of service is guaranteed for all transit
passengers, regardless of their O-D locations. Second, this strategy is
easy to implement, as the public can be informed of the service pol-
icy and what they should expect before using the service. Further-
more, stop-skipping information can be displayed at a bus stop if an
electronic device is installed at the stop or it can be announced by bus
drivers so that passengers whose destinations will be skipped do not
board the wrong bus. This operating method therefore would cause
much less confusion than the general stop-skipping strategy.
With this assumed operating strategy, the route model can be for-
mally dened as follows. Let the bus that is currently waiting at the
terminal for a decision on its stop-skipping pattern for its immedi-
ate trip be denoted Bus 1, as indicated in Figure 1. Because of the
assumed operating strategy, the preceding bus (Bus 0) and the fol-
lowing bus (Bus 2) will have a nonexpressing tripthat is, they will
not be allowed to skip any stops. Because the performance of the cur-
rent bus (Bus 1) and the following bus (Bus 2) will depend on the state
of the preceding bus (Bus 0) as well as the decision on the skipping
pattern of Bus 1, one needs to consider all three buses in formulating
the problem.
The following notations are used in describing the proposed model:
i index of service vehicles, i 0, 1, 2;
j index for identifying the stations or stops on the transit
route, j 1, 2, . . . , N;
r
j
running time between Stop j 1 and j, assumed to be
constant, j 2, . . . , N;
D
i, j
departure time of Bus i at Stop j, i, j;
A
i, j
arrival time of Bus i at Stop j, i, j;

i, j
dwell time of Bus i at Stop j, i, j;
H
i, j
departure headway between Bus i 1 and Bus i at Stop j,
i 1, 2, j;
W
i, jk
number of passengers waiting for Bus i and traveling from
Stop j to Stop k, i 0, 1, 2; 1 j < k N;
L
i, jk
number of passengers traveling from Stop j to Stop k
skipped by Bus i, i 0, 1, 2; 1 j < k N;
L
i, j
number of passengers at Stop j skipped by Bus i, i 0, 1,
2; j 1, . . . , N 1 (note: );
U
i, j
number of passengers boarding Bus i at Stop j, i 0, 1, 2,
j 1, . . . , N 1;
L L
i j i jk
k j
N
, ,

1
Stop 1
(Terminal)

Bus 0
Bus 1 Bus 2
Stop N
Stop 2
GPS
Dispatch
Center
FIGURE 1 Route model. (GPS Global Positioning System.)
V
i, j
number of passengers alighting Bus i at Stop j, i 0, 1, 2,
j 2, . . . , N;
b average boarding time per passenger, a constant;
a average alighting time per passenger, a constant;
average bus acceleration plus deceleration time, a con-
stant;

j, k
average passenger arrival rate at stop j whose destination
is Stop k, 1 j < k N;

j
average passenger arrival rate at stop j (note:
c
1
unit time value associated with passenger waiting time
($/h);
c
2
unit time value associated with passenger in-vehicle time
($/h);
c
3
unit time value associated with vehicle operation time
($/h); and
y
i, j
decision variables to indicate stop status of Bus i at Stop
j. y
i, j
takes two values: y
i, j
1 if Bus i makes Stop j; y
i, j

0, otherwise.
System State Equations
Transit vehicles operating on a given route follow an almost identi-
cal process: they arrive at a stop, dwell at the stop for passengers
boarding and alighting, and then depart for the next stop. This
process starts at Stop 1 and ends at Stop N. The following equations
can be obtained for the relationships between the states of the three
vehicles at individual stops:
Equation 1 indicates that the arrival time of vehicle i at stop j (A
i, j
)
is equal to its departure time at stop j 1 (D
i, j1
) plus the running time
between the two stops plus time lost in acceleration and deceleration.
Equation 2 relates departure time to arrival time and dwell time. Equa-
tion 3 states that the departure headway of Bus i at a stop is the dif-
ference in departure times between itself and the preceding Bus i 1
at the stop, assuming passing is not allowed or at least will not occur
for the three buses under consideration. Equation 4 estimates the bus
dwell time at each stop based on the number of passengers who will
board and alight at the stop, denoted by U
i, j
and V
i, j
, respectively.
It should be noted that the preceding dynamic equations assume
that vehicles will not pass each other over the planning horizon. This
assumption has been used in all existing models and its implication
requires future research. However, it should be pointed out that the
issue of passing and its consequence are less a problem in this model
than in some existing models. This is because this model is expected
to be implemented in a rolling-horizon optimization framework in
which optimal decisions are applied only to those buses that need a
decision at the time the optimization process is invoked. If the opti-
mal stop-skipping decision does lead to passing, it should be reected
in the next round of optimization.
Two initial conditions need to be provided with Equations 1 to 4:
the departure times of Bus 0 at all stops, D
0, j
, for j 1, . . . , N, and

i j i j i j
b U a V i j N
, , ,
, , , . . . , ( ) + 1 2 2 4
H D D i j N
i j i j i j , , ,
, , , . . . , ( )
1
1 2 2 3
D A i j N
i j i j i j , , ,
, , , . . . , ( ) + 1 2 1 2
A D r y y i j N
i j i j j i j i j , , , ,
, , , . . . , ( ) + + +
1 1
2 2
1 2 2 1


j j k
k
N

,
);
1
50 Paper No. 03-3697 Transportation Research Record 1857
the departure times of all buses at Stop 1, D
i,1
for i 0, 1, and 2. The
former is either known (when Bus 0 has already passed the stop at
the time of dispatching Bus 1) or can be predicted by the control cen-
ter based on Bus 0s current location and traffic conditions. In this
study, a simple model was used to predict the estimated time of
arrival based on current vehicle location (which could be obtained
with automatic vehicle location technology), average travel speed,
and average passenger arrival rates. It is also assumed that buses
departure times at the terminal follow scheduled dispatch headway
and their earliest available times at the terminal (after nishing their
previous trips and having a minimum amount of layover time).
The number of passengers boarding and alighting a bus can be
estimated with the following recursive equations:
Equation 5 indicates that the expected number of passengers who
will board Bus i at Stop j (assuming Bus i stops at Stop j) depends on
the number of passengers traveling between Stops j and k (k > j) and
whether the bus will stop at Stop k. Similarly, Equation 6 indicates
that the expected number of alighting passengers for Bus i at Stop j
(assuming Bus i stops at Stop j) depends on the number of passengers
traveling between Stops k and j (k < j) and whether the bus will make
Stop k. The number of passengers waiting for Bus i at Stop j whose
destination is Stop k depends on the number of passengers skipped by
Bus i 1 at Stop j, L
i1, jk
, and the average number of passengers who
arrive at Stop j after Bus i 1 leaves Stop j. Equation 8 species that
the number of passengers destined for Stop k who are stranded by Bus
i 1 at Stop j, L
i1, jk
, will be 0 if Bus i 1 stops at Stops j and k but
otherwise will equal the number of passengers waiting for Bus i 1
at Stop j who have Stop k as their destination.
Application of Equations 7 and 8 requires initial conditions for the
number of passengers skipped by Bus 0 at Stop jthat is, L
0, jk
.
Because Bus 0 is not allowed to skip any stops and because capacity
is assumed not to be a restrictive factor, L
0, jk
0 for j, k 1, . . . , N.
Furthermore, it is assumed that there will be no passengers boarding
at Stop N and alighting at Stop 1that is, U
i, N
0 and V
i, 1
0.
Optimization Model
Transit operations control problems can be formulated in many dif-
ferent ways, depending on the choice of performance criteria and
operating constraints considered. The most commonly used objec-
tive function aims to minimize the total passenger waiting time plus
a discounted amount of the delay to onboard passengers. Because
one of the benets that are expected from the skipping control pro-
posed in this study is a reduction in bus trip time, bus trip time is
included in the objective function. In particular, this model is for-
mulated to minimize the equivalent total cost of passenger waiting
time and passenger in-vehicle time as well as vehicle travel time
subject to the previously formulated system state equations, recur-
sive relationships, initial conditions, and variable restrictions. Stated
mathematically this yields the following model:
L W W y y i j N
i jk i jk i jk i k i j , , , , ,
, , , . , ( ) 1 2 1 8 . .
W L H i j N
i jk i jk j k i j , , , ,
, , , . . . , ( ) +
1
1 2 1 7
V y W y i j N
i j i j i kj i k
k
j
, , , ,
, , , . . . , ( )

1
1
1 2 2 6
U y W y i j N
i j i j i jk i k
k j I
N
, , , ,
, , , . . . , ( )
+

1 2 1 1 5
subject to Equations 1 to 8.
The rst term of the objective function includes two components.
The rst component, (U
i, j
L
i1
, j)H
i, j
/2, computes the total waiting
time of the passengers who arrive after the departure (or passing) of
Bus i 1 at Stop j, assuming random arrival with an average pas-
senger waiting time equal to half the headway. The second compo-
nent represents the total waiting time of those passengers who have
been stranded by Bus i 1 (L
i1
, j) and have to wait for an average
amount of time equal to (H
i1,
j/2 + H
i, j
).
The second term in the objective function calculates the total in-
vehicle time of passengers summed over all O-D pairs. The nal term
computes the total bus trip time. Because time values associated with
passengers and transit vehicles are not of equal importance, these
terms are converted to common units of cost in dollars with the
weighting factors c1, c2, and c3 for passenger waiting time, in-
vehicle time, and bus trip time, respectively. Constraint 10 speci-
fies that stops 1 and N are not to be skipped by Bus 1 or 2, and
Constraint 11 imposes a further no-stop-skipping policy on Bus 2.
Solution Method
The problem formulated in the previous section is a nonlinear 0, 1
programming problem, which can be solved with nonlinear opti-
mization techniques. In this study, because it is necessary only to
decide the stop-skipping pattern of one bus, the problem scale is rel-
atively small. As a result, it was decided to use an exhaustive search
method for optimally solving the problem. The algorithm complex-
ity is exponentialthat is, on the order of 2
N2
, where N 2 is the
number of intermediate stops on the route. Experiments on a set of
realistic cases suggested that this complexity was acceptable for the
simulation analysis. Cases with one-way 14 stops (see Sensitivity
Analysis section) were simulated on a 1.5-GHz personal computer
and it was found that the simulation speed was three to ve times
faster than real-time speed. This algorithm was integrated into a sim-
ulation model for evaluating the effectiveness of the proposed
model, as detailed in the following section.
SENSITIVITY ANALYSIS
The stop-skipping optimization model discussed in the previous sec-
tion was established on the basis of a number of assumptions such as
deterministic travel time and constant headway. From a theoretical
point of view, systems with this stop-skipping control strategy should
always outperform those without this control. However, it is unclear
what magnitude of benets could be expected from application of this
y j N
j 1
0 1 2 1 12
,
, , . , ( ) { . .
y j N
j 2
1 2 1 11
,
, , . , ( ) . .
y y i
i i N , ,
, , ( )
1
1 1 2 10
min
, ,
,
,
,
,
, , ,
Z c U L
H
L
H
H
c U r y
i j i j
i j
i j
i j
i j j
N
i
i jk f i f i f
f j
k
j
N
j
( )
+
|
.

`
,

+
|
.

`
,

]
]
]
+ + + ( ) [ ]

'

'


+ +


1 1
1
1
1 1
2
2
1 1
2 2




+ + + ( ) [ ]
1
1
1
2
3
2 1
2
9
N
i
j i f i j
j
N
i
c r y
, ,
( )
Fu et al. Paper No. 03- 3697 51
strategy and under what conditions this strategy is most benecial.
The objective of this section is to shed some light on these two issues
through a sensitivity analysis with results from a simulation model.
The simulation model used is called SimTransit, which was
developed specically for modeling bus operations under a variety
of operating conditions and dispatch controls (17, 18). The model
includes three main components: a dispatch module, a traffic simu-
lator, and a geographic information systems-based animator. The
dispatch module is a representation of a transit dispatch center, inte-
grating functions such as service monitoring, state prediction, and
dispatching. The module was modied in this study to include the
skipping optimization model, which determines the optimal skip-
ping pattern based on the estimated times of arrival of individual
buses under consideration and the expected passenger demand.
The sensitivity analysis was performed on a real-life bus route
(Route 7D) operated by Grand River Transit (GRT), Regional Munic-
ipality of Waterloo, Ontario, Canada. Route 7D is located in the twin
cities of Kitchener and Waterloo, which have a combined population
of 293,800 within an area of 203 km
2
. Route 7D includes 28 major
stops, starting from the Transportation Centre (TC) located in down-
town Kitchener, via the University of Waterloo (a major O-D), and
back to the TC terminal, as indicated in Figure 2. The original head-
way is 7.5 min. The afternoon peak period for this route was used (the
base demand prole for this route is presented in Figure 3). These data
were partially provided by GRT and partially collected by University
of Waterloo students around May 2001. Average passenger board-
ing and alighting times were assumed to be 4 and 2 s per passenger,
respectively. A total of 15 buses were used and each bus was dis-
patched from the TC terminal at a headway of 5 min. A uniform devi-
ation of t90 s from the scheduled headway was introduced to model
the inherent variation in dispatching headway. To model bus travel
time variation, a normal distribution with a coefficient of variation
(COV) of 0.20 was assumed for all links along the route. The assumed
travel time COV (0.20) represents the typical variation observed in
the eld. Two stops were selected as the control points: the TC ter-
minal (Stop 1) and the University of Waterloo station (Stop 14). Val-
ues of $20/h, $10/h, and $50/h were used for the objective function
weighting factors c1, c2, and c3, respectively.
Each simulation run generates statistics on the following four mea-
sures of effectiveness: (a) passenger waiting time, (b) passenger in-
vehicle time, (c) bus travel time, and (d) total weighted cost. These
measures of effectiveness are used in the following sensitivity analy-
sis on three model parameters: passenger demand, headway, and
travel time. For each parameter setting, bus operations with and with-
out skipping controls were simulated and relative reductions in these
four performance measures were used for comparison. In the simu-
lation, the rst hour was treated as a warm-up period for the total
6-h simulation run.
Sensitivity to Passenger Demand
To determine how the effectiveness of the proposed control strategy
depends on the level of passenger demand or how the proposed strat-
egy would perform during different times of day, cases with four
levels of O-D demand, including base demand, 1.5 base demand,
1.8 base demand, and 2.0 base demand, were simulated. The
simulation results are presented in Figure 4. As expected, the total
weighted cost, which was to be minimized explicitly in the under-
lying optimization process, improved under all demand scenarios.
The magnitude of the improvement, however, depended on the level
of passenger demand. At low passenger demand (base case), the total
52 Paper No. 03-3697 Transportation Research Record 1857
As passenger demand increases, the reduction in passenger wait-
ing time, in-vehicle time, and bus trip time also increases. This trend
continues until the demand reaches a certain level, beyond which
both passenger waiting time and in-vehicle time start to decrease,
compared with bus travel time. As a result, there is an optimal level
of demand at which the total benet or reduction in total weighted cost
is maximized. This peaking phenomenon can also be attributed to
the two conflicting effects of the stop-skipping strategy discussed
previously.
Sensitivity to Bus Travel Time Variation
Variation in bus travel time is another factor that causes buses to
deviate from their scheduled headways. This variation can cause
buses to bunch and to run behind schedule. The degree of variation
in travel time therefore should have some impact on the effectiveness
of the proposed stop-skipping control strategy, which was intended
to prevent bunching and reduce lateness. Intuitively, the higher
the travel time variability, the higher the bus headway variability,
and thus the more opportunities there are for applying stop-skipping
controls. However, higher travel time variation also means larger
errors in the estimated times of arrival used as input to the optimiza-
tion model. This could result in suboptimal scheduling solutions.
This intuitive observation is supported by the simulation results pre-
sented in Figure 5, where the curves represent the relationship
Transportation
Center
University of
Waterloo
0
50
100
150
200
250
P
a
s
s
e
n
g
e
r

D
e
m
a
n
d

(
p
e
r
s
o
n
/
h
o
u
r
)
Boarding
Alighting
Transportation
Center
(Origin)
University of
Waterloo
(Midstop)
Transportation
Center
(Terminus)
FIGURE 2 Simulation model interface with an example route.
FIGURE 3 Passenger demand profile.
combined benet was relatively small (<1%). In fact, the implemen-
tation of skipping control had the negative effect of increasing pas-
senger waiting time. This is because of the two opposing effects the
stop-skipping controls have on passengers. On the one hand, stop
skipping will increase the waiting time of those passengers whose
O-D stops are skipped. On the other hand, allowing buses to skip
some stops can prevent or mitigate bus bunching and thus reduce pas-
senger waiting time. At a low level of passenger demand, bus bunch-
ing is less likely to develop, and therefore the effect of increasing
passenger waiting time is more likely to be dominant.
between the four measures of effectiveness and the variability of link
travel time. The high-demand scenario of 1.8 base demand was
used in this analysis and the variability of link travel time was dened
by the COV or the ratio of standard deviation to mean.
For the case simulated, the stop-skipping control is most effective
when the coefficient of variation of travel time is close to a critical
value of 0.20. When the travel time variability is smaller than the crit-
ical value, the control strategy is still benecial but with reduced ben-
ets. The benets decrease quickly when the variability goes beyond
the critical value, suggesting that cautions must be taken when apply-
ing the control model to highly varied traffic conditions, as it could
have no effect at all or, even worse, be countereffective.
Sensitivity to Headway
When a stop is skipped, the passengers who are waiting at the stop
must wait for the next bus and consequently incur the additional
waiting time of one headway. Intuitively, the smaller the operating
headway is, the smaller the effect skipping controls will have on
these passengers. To gauge the magnitude of this effect, the 1.8
base demand scenario was simulated with bus operating headway
Fu et al. Paper No. 03- 3697 53
varied at ve levelsnamely, 3, 5, 7, 10, and 15 min. The demand
corresponding to the 5-min headway was set at 1.8 base demand,
which was then used as a base case to determine the level of demand
for other headways using the scaling factor 5/h, where h is the head-
way under consideration. For example, for the case with a headway
of 10 min, a scaling factor of 0.5 ( 5/10) was used to calculate the
demand. The logic behind this is that headways for transit routes in
high-demand corridors usually are determined on the basis of pas-
senger demand. The higher the travel demand is, the lower the
headway usually will be. The simulation results for this part of the
study are presented in Figure 6.
As expected, the benets of the stop-skipping control decrease
monotonically as the operating headway increases. This pattern sug-
gests that the proposed control is more appropriate for routes with a
short headway than for those with a long headway. The total benets
approached 0 when the bus headway was increased to 10 min or more.
Combined Control
Past studies have suggested that stop-skipping control may be applied
as a complement to another more popular bus control strategy
-2%
0%
2%
4%
6%
8%
10%
1 1.5 1.8 2
Passenger Demand (x Base Demand)
R
e
d
u
c
t
i
o
n
Passenger Waiting Time
Passenger In-Vehicle Time
Bus Travel Time
Weighted Total
-10%
-8%
-6%
-4%
-2%
0%
2%
4%
6%
8%
10%
- 0.10 0.20 0.30 0.40
Link Speed Deviation
R
e
d
u
c
t
i
o
n
Passenger Waiting Time
Passenger In-Vehicle Time
Bus Travel Time
Weighted Total
FIGURE 4 Control effectiveness versus passenger demand.
FIGURE 5 Control effectiveness versus link speed variations.
namely, holding control (8, 14). The objective of holding control is to
purposely delay those buses that are ahead of their planned schedule
or too close to preceding buses. Conversely, stop-skipping control is
used to speed up those buses that are late or too far from the preced-
ing buses. To study this effect, a simulation was run that combined
holding and skipping control strategies and compared the results with
those in which only holding or skipping strategies were applied. The
results are presented in Table 1. It can be observed that the holding
plus skipping control has indeed improved the system performance
compared with the skipping or holding control applied in isolation.
This suggests that the negative effect on bus travel time caused by
the holding control has been compensated by incorporating the
stop-skipping strategy.
CONCLUSIONS AND FUTURE RESEARCH
In this paper, a new bus operations control strategy has been pro-
posed in which stop skipping is applied to every other bus dispatched
from the terminal. The novelty of this control strategy is that a min-
imum service frequency can be ensured at all stops while both pas-
sengers and transit agencies can still enjoy the benet of reduced
travel time and operating costs. The underlying control problem was
formulated as a nonlinear 0, 1 programming problem and solved
through an exhaustive search process. A simulation model was used
in a sensitivity study to investigate the impacts of changes in various
operating conditions such as demand, travel time, and headway on
the effectiveness of the control strategy. The analysis has provided
the following insights:
54 Paper No. 03-3697 Transportation Research Record 1857
1. Stop-skipping control is an effective strategy to improve transit
service quality and operating efficiency. Passenger in-vehicle time,
waiting time, and operation vehicle trip time can be reduced in a wide
range of operating conditions.
2. Stop-skipping controls are most effective on those bus routes
with high passenger demand and short headway.
3. Stop-skipping controls should be used only on those routes
with an appropriate range of travel time variation. Routes with travel
time variation that is too low or too high may not benet from this
strategy.
4. A stop-skipping strategy can be applied in combination with
other controls such as a holding control for further improvement in
system performance.
The work presented in this paper is by no means complete and fur-
ther research is needed in the following directions. First, faster algo-
rithms need to be developed to replace the currently implemented
enumeration method if large-sized problems are to be solved in real
time. Second, more accurate prediction models should be developed
to truly take advantage of real-time information on bus location, travel
time, and passenger counts. Lastly, integration with other controls
such as real-time deadheading and short turning should be explored
to maximize the potential of the proposed strategy.
ACKNOWLEDGMENT
This research was supported by the Natural Sciences and Engineer-
ing Research Council of Canada.
0%
1%
2%
3%
4%
5%
6%
7%
8%
9%
10%
3 min 5 min 7 min 10 min 15 min
Headway

R
e
d
u
c
t
i
o
n
Passenger Waiting Time
Passenger In-Vehicle Time
Bus Travel Time
Weighted Total
Control
Strategy
Reduction in
Average In-Vehicle
Time
Reduction in
Average Waiting
Time
Reduction in
Average Bus
Trip Time
Reduction in
Weighted
Total
Skipping 2.84% 8.91% 4.66% 5.62%
Holding 6.46% 30.72% -0.67% 16.17%
Holding +
Skipping
8.63% 32.56% 0.58% 18.11%
FIGURE 6 Control effectiveness versus headway.
TABLE 1 Effects of Holding and Stop-Skipping Controls
REFERENCES
1. Furth, P. G., and F. B. Day. Transit Routing and Scheduling Strategies
for Heavy-Demand Corridors (Abridgment). In Transportation Research
Record 1011, TRB, National Research Council, Washington, D.C., 1985,
pp. 2326.
2. Turnquist, M. A. Zone Scheduling of Urban Bus Route. Journal of
Transportation Engineering, Vol. 105, No. 1, 1979, pp. 113.
3. Turnquist, M. A. Strategies for Improving Reliability of Bus Transit
Service. In Transportation Research Record 818, TRB, National
Research Council, Washington, D.C., 1981, pp. 713.
4. Jordan, W. C., and M. A. Turnquist. Zone Scheduling of Bus Routes to
Improve Service Reliability. Transportation Science, Vol. 13, No. 3,
1979, pp. 242268.
5. Furth, P. G. Short Turning on Transit Routes. In Transportation
Research Record 1108, TRB, National Research Council, Washington,
D.C., 1987, pp. 4252.
6. Vijayaraghavan, T. A. S., and K. M. Anantharamaish. Fleet Assignment
Strategies in Urban Transportation Using Express and Partial Services.
Transportation Research A, Vol. 29, 1995, pp. 157171.
7. Ceder, A. Optimal Design of Transit Short-Turn Trips. In Transportation
Research Record 1221, TRB, National Research Council, Washington,
D.C., 1989, pp. 822.
8. Site, P. D., and F. Filippi. Service Optimization for Bus Corridors with
Short-Turn Strategies and Variable Vehicle Size. Transportation
Research A, Vol. 32, No. 1, 1998, pp. 1938.
9. Furth, P. G. Alternating Deadheading in Bus Route Operations. Trans-
portation Science, Vol. 19, No. 1, 1985, pp. 1328.
10. Eberlein, X. J. Real-Time Control Strategies in Transit Operations:
Models and Analysis. Ph.D. dissertation. Department of Civil and Envi-
ronmental Engineering, Massachusetts Institute of Technology, 1995.
Fu et al. Paper No. 03- 3697 55
11. Eberlein, X. J., N. H. M. Wilson, C. Barnhart, and D. Bernstein. The
Real-Time Deadheading Problem in Transit Operations Control. Trans-
portation Research B, Vol. 32, No. 2, 1998, pp. 77100.
12. Li, Y., and F. Wu. Real-Time Scheduling on a Transit Bus Route.
Computer-Aided Transit Scheduling Proceedings, Montreal, Canada,
Aug. 1990.
13. Li, Y., J.-M. Rousseau, and M. Gendreau. Real-Time Dispatching of
Public Transit Operations With and Without Bus Location Information.
Computer-Aided Transit Scheduling Proceedings, Lisbon, Portugal,
July 1993.
14. Li, Y. Real-Time Scheduling on a Transit Bus Route. Ph.D. disserta-
tion. cole des Hautes tudes Commerciales, University of Montreal,
1994.
15. Araya. S., K. Abe, and K. Fukumori. An Optimal Rescheduling for
Online Train Traffic Control in Disturbed Situations. IEEE, New York,
1983, pp. 489494.
16. Lin, G., P. Liang, P. Schonfeld, and R. Larson. Adaptive Control of
Transit Operations. Final Report for Project MD-26-7002, University
of Maryland, 1995.
17. Yang, X. Transit Control and Simulation Under Real-Time Informa-
tion. Masters thesis. Department of Civil Engineering, University of
Waterloo, 2002.
18. Fu, L., and X. Yang. Design and Implementation of Bus-Holding
Control Strategies with Real-Time Information. In Transportation
Research Record: Journal of the Transportation Research Board,
No. 1791, TRB, National Research Council, Washington, D.C., 2002,
pp. 612.
Publication of this paper sponsored by Committee on Transportation Network
Modeling.